Assessment of Intelligibility and Comprehensibility
Last Updated: 9/13/04 10:53 AM
There are many factors that influence speech intelligibility. These
factors may reside within the speaker, the listener, the context, or
the environment. Because of the broad nature of this concept, we often
prefer the more inclusive term of comprehensibility
(reference for article on comprehensibility).
When we assess intelligibility, we do so at the word,
sentence, and discourse level.
Supplemental assessment can also be conducted through instrumentation. For example, the Computerized Speech Lab (CSL) can provide information about pitch and loudness. In addition, we use a particular program called the Motor Speech Profile available on the CSL.
Word Level Assessment
At the word level we look for particular patterns of errors that may
provide information about underlying phonological processes. Extensive
errors may reflect a poorly developed phonological system. They may
also reflect underlying physiological deviations such as inappropriate
modulation of voice onset or insufficient control of air expenditure.
The FSST is a test that is particularly valuable for assessment when such underlying deviations occur at this level.
We have found that describing errors to students using a distinctive
feature approach with traditional notions of place, manner, and voicing
is valuable, particularly as an instructional tool. For example, distinctive
feature terminology is valuable when describing visual information on
a speech spectrogram (example video – MG and D).
This is demonstrated in the technology section.
Older students tend to appreciate this descriptive information as it
enhances their metacognitive understanding of speech production. Primarily
for this reason, we have continued to use the Fisher-Logemann
Test of Articulation Competence (ref) which has a scoring sheet
that partitions errors into place, manner, and voicing categories.
Identifying errors and discerning patterns and processes is a critical
step in the assessment process. Some speech samples are presented here
to illustrate a variety of phonetic errors and the acoustic effects
of poorly coordinated respiration, phonation, and articulation.
[Click on each of the following words to hear a sample]
Example 1
These words show difficulty managing voicing. Notice that phonemes are
voiced in initial and medial position of words but devoiced in final
position; notice also the extraneous voicing following glottal plosives
pin box bib finger before nothing mouth top bed dishes chair coat egg
Example 2
These words show difficulty controlling air for fricative and plosive
manner of production; notice also the difficulty controlling the onset/offset
of voicing
water finger knife valentine five thumb mouth letter ladder bed seal
glasses bus size
Example 3
These words show nasal substitutions for initial sounds with different
manners of production; notice correct production for /m/ and /n/
man hammer drum valentine thumb top dog nose money pen seal coat
Example 4
These words show a substitution of a plosive manner of production for
nasals and glides
man hammer drum valentine thumb leaf balloons nose money pen yellow
beyond rain ring behind
Example 5
Notice how control of air expenditure and glottal and velar variations
influence the production of these words
water finger top letter dog tail dozen dishes cage yellow carrot car
coat hanger
Example 6
Notice how vocal tension interferes with the coarticulatory aspects
of speech production
Paper baby water man hammer valentine smooth dog balloons nose glasses
zebra dozen shoe
Example 7
Notice the elevated pitch and resonance changes that occur with certain
words; follow-up assessment should determine whether such changes are
systematic or random
Box drum valentine bed tail
Example 8
Notice how variations may reflect native or first language phonology
paper water flower hammer drum television letter ladder balloons zebra
garage rain carrot car ring
Sentence Level
We also assess intelligibility at the sentence level. One type of analysis
looks at whether there is consistency in production at the word and
sentence level. Another analysis looks at intelligibility as a function
of how well a listener can understand words produced in context. This
is a write-down intelligibility measure that utilizes Clark sentences
(link to Clark reference).
A student is recorded reading a corpus of ten sentences and a listener
reviews the recording and orthographically transcribes what he/she hears.
The number of key words correctly understood is computed and transformed
to an intelligibility rating (link to Metz & Samar article on intelligibility).
The intelligibility scale ranges from 1 to 5, with 1 indicating speech
that is unintelligible.
Samples of sentences reflecting this range of intelligibility are available
for listening. There are three for each rating. You may want to listen
to one sentence at a time and, after listening, view the target.
Click to Hear Sample Click to see target sentence
• 1.0 Rating
? These target sentences will appear after the student click (or points) to see the target sentence:
1. sentence one
2. sentence two
3. sentence three
• 1.5 Rating ? 1. sentence one
2. sentence two
3. sentence three
• 1.8 Rating ? 1. sentence one
2. sentence two
3. sentence three
• 2.1 1. sentence one
2. sentence two
3. sentence three
• 2.6 1. sentence one
2. sentence two
3. sentence three
• 2.9 1. sentence one
2. sentence two
3. sentence three
• 3.0 1. sentence one
2. sentence two
3. sentence three
1. sentence one
2. sentence two
3. sentence three
• 3.6 1. sentence one
2. sentence two
3. sentence three
• 3.9 1. sentence one
2. sentence two
3. sentence three
• 4.0 1. sentence one
2. sentence two
3. sentence three
• 4.4 1. sentence one
2. sentence two
3. sentence three
• 4.7 1. sentence one
2. sentence two
3. sentence three
• 4.4 1. sentence one
2. sentence two
3. sentence three
• 5.0 1. sentence one
2. sentence two
3. sentence three
• 5.0 1. sentence one
2. sentence two
3. sentence three
Discourse Level
Intelligibility and comprehensibility are assessed at the discourse
level through analyses of read and conversational speech. Comprehensibility
is influenced by segmental and suprasegmental aspects of speech production
and by all aspects of language including syntactic adequacy and complexity
and semantic variation and appropriateness. In addition, pragmatic
aspects of language production influence the comprehensibility of
a message, particularly in conversational interactions.
Speech intelligibility can be assessed through an evaluation of read
discourse. We tend to use The Rainbow Passage for this purpose. This
task eliminates language formulation and competence factors and focuses
attention on the ongoing coordination of respiration, phonation, and
articulation in continuous speech. However, it also introduces an
oral reading factor. Read speech can be evaluated for intelligibility
using a five-point rating scale, with 1 indicating unintelligible
speech and 5 indicating fully intelligible speech. It should be noted
that research conducted by Metz, Samar and Schiavetti has found that
the write down intelligibility procedure mentioned above yields a
more valid and reliable measure of intelligibility.
The suprasegmental aspects of continuous discourse are evaluated
using the NTID Speech and Voice Evaluation Form. This can be done
on read and/or spontaneous speech samples. Comparing performance on
both tasks can be informative. {Speech and Voice form PDF?}
Listen to these examples.
- In this example, overall tension and instrusive voicing distracts from intelligibility {SS Gav.._L_03_04.WAV} - need transcript
In the second, she is spontaneously relating the same story. Note that intelligibility is comparable but spontaneous speech is fater and has more pitch and loudness breaks {SR two speech samples}
You can learn more about rating speech and voice parameters by using the self-instruction module.
More information on collecting and analyzing spontaneous speech samples
is included in Language Assessment
